Pith. sign in

REVIEW 4 major objections 6 minor 49 references

CoLa: Chinese Character Decomposition with Compositional Latent Components

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read CoLa learns Compositional Latent components of Chinese characters without human-defined decomposition and uses them for zero-shot Chinese character recognition.

desk verdict Large zero-shot gains from slot-attention decomposition, but the key ablation (is it the compositionality or just the teacher features?) is missing. read the letter →

arxiv 2506.03798 v1 pith:7362YZ6D submitted 2025-06-04 cs.CV

Fan Shi , Haiyang Yu , Bin Li , Xiangyang Xue This is my paper · ORCID
classification cs.CV
keywords Chinesecharacterrecognitionzero-shotcompositionallatentcomponentsslotattentionradicaltemplatematchingvariationalinferenceoraclebonecharacters
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper introduces CoLa, a deep latent variable model that learns to split Chinese character images into a small number of latent components without using radical or stroke labels. The central claim is that these learned components act as a shared alphabet: for a previously unseen character, CoLa encodes both the input image and printed-font templates into the same latent-component space, and the class is chosen by matching the input's components against each class's template components. The authors report that this beats earlier zero-shot Chinese character recognition methods, with 68.59% versus 21.79% for CCR-CLIP on the 500-class HWDB handwritten setting and an average accuracy gain of about 60% in radical zero-shot settings. If correct, the model shows that decomposition-and-recombination for recognizing new characters can be learned directly from data and can transfer across writing styles, historical documents, and even oracle bone scripts.

What carries the argument

The machinery is slot attention with a shared component order, plus template matching. Slot attention is an iterative attention mechanism that groups image features into $K$ slots; here $K=3$, and each slot becomes a compositional latent component. A frozen teacher encoder's visual features are the reconstruction target, so the slots must encode structure rather than low-level pixels. Templates are encoded with the same slot-attention encoder, and class prediction treats each character class as a mixture of Gaussians whose mean is the average of its template components, which turns zero-shot recognition into a latent-space comparison. The fixed observed variable $\epsilon$, which initializes the slots in the same order for all images, is what makes component $k$ comparable across styles and across the printed-to-handwritten domain gap.

What would settle it

Run the 500-class HWDB character zero-shot experiment with a random decomposition order sampled independently for each image at test time; if accuracy stays near the reported 68.59%, the fixed order is not load-bearing, while a large drop would confirm that slot alignment is the mechanism. A complementary check is to permute the three template components at test time and see whether recognition and retrieval accuracy fall.

Watch

Extended reading notes

Core claim

CoLa's central discovery is that slot attention, trained by reconstructing frozen teacher features and by predicting classes through latent template matching, discovers compositional components of Chinese characters that are consistent enough across images to support zero-shot recognition. The components are not human-defined radicals, yet they are repeatable and structurally meaningful: with three slots, each component attends to a distinct region of the character. A fixed global component order $\epsilon$ aligns slot $k$ across every input and template image, so recognition can compare component $k$ of an unseen handwritten character with component $k$ of printed templates. The class decision is a nearest-class rule in latent space, where each class is a Gaussian mixture component centered at the mean of its $N$ template component vectors. On the paper's experiments, this unsupervised decomposition outperforms prior radical- and stroke-based methods in both character and radical zero-shot settings, and the same model trained on historical documents decomposes oracle bone characters without retraining.

Load-bearing premise

The load-bearing premise is that the fixed global component order keeps slot $k$ aligned across every input and template image, so comparing component $k$ between a handwritten input and a printed template is semantically meaningful; if that alignment fails, the class prediction compares mismatched components and accuracy collapses.

Editorial extensions

If this is right

  • Human-defined radical or stroke annotations are not required for zero-shot Chinese character recognition; the components are discovered from data.
  • Recognition is a fast template-matching operation in latent space, avoiding the slow autoregressive decoding of radical or stroke sequences (the paper reports 9 ms versus 1666 ms per batch for DenseRAN).
  • The same decomposition transfers to unseen scripts: a model trained only on historical documents parses oracle bone, Japanese, and Korean characters and retrieves visually similar samples.
  • Rare or unseen radicals do not break recognition because the model's components are tied to visual structure, not to a predefined radical inventory.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper, a stronger test would be cross-writer or cross-era style shift, where the printed-to-handwritten gap is replaced by an unseen handwriting style gap; the fixed-order alignment has not been tested there.
  • Because the decoder turns each component into a mask and feature map, the learned components could be reused for generation or style transfer, not just recognition.
  • If the fixed-order assumption is what matters, then slot permutation at test time should destroy accuracy; that experiment would separate semantic components from positional slots as the actual carrier of generalization.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper proposes CoLa, a slot-attention-based deep latent variable model that decomposes Chinese character images into K latent components without radical or stroke supervision. The components are decoded to reconstruct teacher features (a frozen DINOv2 encoder plus a trained CNN), and classification is performed by comparing the latent components of the input against per-character template prototypes through a mixture-of-Gaussians rule (Eq. 7). Experiments report large zero-shot gains over prior radical-, stroke-, and CLIP-based methods on handwritten (HWDB), printed-font, and historical-document benchmarks, along with qualitative component visualizations and cross-dataset retrieval experiments on oracle bone characters.

Significance. If the reported gains are robust and genuinely attributable to the compositional latent-component mechanism, CoLa would be a significant step toward practical zero-shot Chinese character recognition that avoids human-defined decomposition schemes. The paper has notable strengths: it reports standard deviations over three trials for its own method (Appendix D.2), makes the template-matching inference rule explicit, and includes a component-order experiment (Appendix D.3) that addresses slot-alignment in a limited way. The central empirical claim is, however, currently underdetermined by the absence of controls that isolate the compositional mechanism, and the printed-font evaluation has a potential template/test overlap. Most of these concerns are addressable with additional experiments, so the paper merits revision rather than rejection.

major comments (4)
  1. [Sec. 4.2, Table 1; Eq. (10)] The central claim that the reported gains are due to compositional latent components is underdetermined. CoLa's recognition rule (Eq. 7) is a nearest-prototype classifier in the slot-latent space, and the only learning signal for S is reconstruction of the teacher features F (Eq. 10), where F comes from a frozen DINOv2 plus a CNN trained on the training classes. The paper provides no K=1 ablation, no control that matches F (or averaged F) directly to templates without slot decomposition, and no ablation that replaces the DINOv2 teacher with a randomly initialized or weaker encoder. Without these controls, the large improvements in Table 1 are equally consistent with a strong-teacher-plus-template-matching explanation in which the slot-attention factorization is not the active ingredient. Please add at least a K=1 variant and a direct teacher-feature matching baseline.
  2. [Appendix B; Sec. 4 dataset construction] The Printed benchmark may be contaminated by template-font overlap. The Printed dataset is generated from 105 font files, and the template set is generated from 'commonly used printed fonts' (Appendix B). The paper does not state whether the fonts used to render the test images are disjoint from the fonts used to generate templates. If they are not disjoint, the Printed rows in Table 1 (e.g., 78.10 at 500 classes and 92.70 at 2755) largely reflect near-duplicate template retrieval rather than zero-shot generalization. Please report the font split, or rerun the Printed experiments with template fonts excluded from the set of fonts used to generate test images.
  3. [Sec. 3.2; Appendix D.3] The class-prediction term in Eq. 7 compares component k of the input with component k of each template, which requires that the fixed global order epsilon aligns slots semantically across inputs, templates, styles, and the printed-to-handwritten domain gap. The only evidence provided is a small set of visualizations in Appendix D.3. Please add a quantitative assessment of slot correspondence: for example, measure whether randomly permuting slot indices across styles degrades accuracy, or evaluate assignment consistency of the learned components across the N templates of the same character. Without such a check, the nearest-component comparison in Eq. 7 may be based on mismatched components.
  4. [Appendix A.4, Eq. (19)] The paper acknowledges in Eq. 19 that the two KL regularizers are identically zero because the variational distribution and the generative process share the same encoder. This means the ELBO reduces to an expected reconstruction plus an expected classification term, with no posterior regularization; the 'deep latent variable model' framing is therefore not a standard variational autoencoder. The authors should either provide a principled reason for keeping the sampling step and the Gaussian likelihood, or reframe the method as a deterministic encoder-decoder with a teacher-feature reconstruction objective and a classification loss. This issue does not by itself invalidate the empirical results, but it affects the theoretical novelty claim.
minor comments (6)
  1. [Abstract] The phrase 'outperforms previous methods in both character the radical zero-shot CCR' is ungrammatical; it should be 'in both character and radical zero-shot CCR.'
  2. [Sec. 3.1] K is described as the 'maximum number of components in the images,' but in practice it is a fixed number (K=3 in the experiments); the wording should be 'the number of latent components.'
  3. [Appendix A.4, Eqs. (20)-(21)] The numerator in Eqs. (20) and (21) uses \tilde{T}_{c,n}, which appears to be a typo for \tilde{T}_{y,n} as in Eq. (18); please correct it.
  4. [Tables 1 and 2] Tables 1 and 2 report only a single run for all baseline methods; adding standard deviations or stating the number of seeds for the key comparisons would make the comparison fairer, especially given the very large margins claimed.
  5. [Appendix B; Sec. 4.2] The historical-document dataset is not released and its size and class distribution are not given; please provide dataset statistics and a data-release plan to support reproducibility.
  6. [Fig. 8 caption] The caption contains a typo: 'Fix initialization' should be 'Fixed initialization.'

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: zero-shot predictions are retrieval outcomes over template components, not quantities fitted from test labels.

full rationale

CoLa's derivation chain (Sec. 3.2, Eqs. 3, 7, 10) defines the class prediction as a nearest-prototype comparison: p(y=i|S,T) is a softmax over Gaussian distances from the input latent components S to the per-class template component means, and the final label is taken as the class with highest probability. At test time, the templates are font-derived images for unseen classes, the teacher encoder is frozen, and the reported hyperparameters (K=3, N=10, lambda=0.01) do not encode the benchmark accuracies. No fitted parameter is renamed as a prediction, and no equation defines the test accuracy in terms of a fitted quantity. The paper's own prior work appears only as baseline methods (e.g., CCR-CLIP [17], SD [15]); no load-bearing premise is justified by a self-citation chain, and no uniqueness theorem is imported from the authors' earlier work. A legitimate concern is that the paper does not run a K=1 or teacher-feature-only control, so the claim that compositional slot components cause the observed gains is underdetermined; however, underdetermination by missing ablations is an experimental-evidence issue, not circularity. Appendix D.3's component-order experiment addresses alignment but does not test whether compositionality is necessary, and again it does not make the result definitionally equal to an input. The unresolved '[? ?]' and '[? ]' citation placeholders in Appendix A.1 and C are completeness artifacts that do not enter the derivation. Thus no circular step can be exhibited from the paper's own equations or argumentation.

Assumptions & free parameters 4 free parameters · 7 assumptions · 0 invented entities

No new physical entities are postulated. The compositional latent components S and T are learned representations, not invented objects; their existence is evidenced only through visualizations and downstream accuracy.

free parameters (4)
  • K (number of latent components) = 3
    Chosen by hand; controls the granularity of decomposition. The paper notes K encourages interpretable components and sets K=3 for all experiments (Sec 4 Training Details).
  • N (templates per character) = 10
    Chosen by hand; used to estimate class centers in Eq. 7. Larger N would reduce variance but increase compute.
  • lambda (prediction term weight) = 0.01 (warmup: 0)
    Tuned by hand; balances reconstruction vs classification in Eq. 10. The paper warms up with lambda=0 then enables 0.01 (Appendix C).
  • sigma (fixed std of Gaussian likelihoods) = sqrt(2)/2
    Set to simplify the ELBO (Appendix A.4); affects the scale of both reconstruction and classification losses.
assumptions (7)
  • standard math Slot attention iteratively clusters visual features into K components (Locatello et al. mechanism).
    Used in Eq. 4-5 for input and template encoding; assumed to produce meaningful components without component-level supervision.
  • standard math SGVB/ELBO provides a valid lower bound on the log-likelihood.
    Used to justify Eq. 2; the derivation in Appendix A.1 is standard, but with missing citation placeholders.
  • domain assumption Chinese characters can be represented by at most K=3 compositional components.
    The model fixes K=3 for all characters, including complex historical glyphs; no adaptive component count.
  • domain assumption Printed-font template images are representative class centers in the latent space.
    Class prediction uses the mean of N template encodings as each class center (Eq. 7); assumes these centers generalize to handwritten and historical inputs.
  • domain assumption Teacher features (frozen DINOv2 plus trained two-layer CNN) carry compositional structure suitable for reconstruction.
    Feature decoding reconstructs teacher features from slots; if teacher features are not decomposable into K=3 additive components, the learned slots may not be compositional.
  • ad hoc to paper A fixed global component order epsilon aligns slot k across images.
    Sec 3.1 requires all images to follow the same order; Appendix D.3 shows fixed initialization gives consistent order. This alignment is assumed, not proven, and is load-bearing for matching.
  • ad hoc to paper Generative p and variational q share the same encoder, making KL regularizers identically zero.
    Eq. 19 sets R_input and R_temp to 0 by construction; this collapses the VAE objective to reconstruction plus classification and is not a principled variational regularization.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CoLa: Chinese Character Decomposition with Compositional Latent Components." pith.science (2026). https://pith.science/paper/7362YZ6D

@misc{pith2026250603798,
  author       = {Pith},
  title        = {Pith review of: CoLa: Chinese Character Decomposition with Compositional Latent Components},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7362YZ6D}},
  note         = {Machine review of arXiv:2506.03798}
}
read the original abstract

Humans can decompose Chinese characters into compositional components and recombine them to recognize unseen characters. This reflects two cognitive principles: Compositionality, the idea that complex concepts are built on simpler parts; and Learning-to-learn, the ability to learn strategies for decomposing and recombining components to form new concepts. These principles provide inductive biases that support efficient generalization. They are critical to Chinese character recognition (CCR) in solving the zero-shot problem, which results from the common long-tail distribution of Chinese character datasets. Existing methods have made substantial progress in modeling compositionality via predefined radical or stroke decomposition. However, they often ignore the learning-to-learn capability, limiting their ability to generalize beyond human-defined schemes. Inspired by these principles, we propose a deep latent variable model that learns Compositional Latent components of Chinese characters (CoLa) without relying on human-defined decomposition schemes. Recognition and matching can be performed by comparing compositional latent components in the latent space, enabling zero-shot character recognition. The experiments illustrate that CoLa outperforms previous methods in both character the radical zero-shot CCR. Visualization indicates that the learned components can reflect the structure of characters in an interpretable way. Moreover, despite being trained on historical documents, CoLa can analyze components of oracle bone characters, highlighting its cross-dataset generalization ability.

Figures

Figures reproduced from arXiv: 2506.03798 by the authors.

Figure 1
Figure 1. Different types of Chinese character recognition methods. (a) character-based methods extract monolithic representations for prediction; (b) and (c) are stroke-based and radical-based methods requiring human-defined decomposition schemes to predict stroke and radical sequences; (d) the proposed CoLa automatically decomposes characters into compositional components. characters. Inspired by these cognitive mechanisms,… view at source ↗
Figure 2
Figure 2. The probabilistic graphical model of CoLa. Gray nodes represent observed variables, while white nodes represent latent variables. Black solid lines indicate the generative process, and red dashed lines indicate the varia￾tional distribution. The probabilistic graphical model of CoLa is shown in Fig￾ure 2. CoLa decomposes a Chinese character image into components to extract individual representations in the latent sp… view at source ↗
Figure 3
Figure 3. The generative process of CoLa. CoLa extract compositional latent components of the input image and template images in the input encoding anf feature decoding processes, respectively. The input compositional latent components are decoded to reconstruct teacher features. The template compositional latent components constitute a mixture of Gaussians in the latent space, which is used to predict the class of the input … view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Visualization of the compositional latent components. For each example, the left side are compositional latent components (Cmp#1 ∼ Cmp#3), and the right side is the human-defined radical decomposition scheme of the character. The radical regions are highlighted using r…
Figure 5
Figure 5. Figure 5: Visualization results on Oracle Bone Characters (OBCs). Left: we attempt to use CoLa trained on the historical document characters to parse components of OBCs from the HUST-OBC dataset. Right: we select four examples and visualize their top 10 similar samples from 1280…
Figure 6
Figure 6. Figure 6: Visualization of the samples from (a) Printed, (b) HWDB and (c) Historical Documents. [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: Visualization of the compositional latent components on the different datasets [PITH_FULL_IMAGE:figures/full_fig_p019_7.png]
Figure 8
Figure 8. Figure 8: The compositional latent components learned with different orders. (a) Random initial￾ization. The components in each panel are learned with randomly sampled ϵ. (b) Fix initialization. The components in each panel are learned using a fixed ϵ. 20 [PITH_FULL_IMAGE:figur…
Figure 9
Figure 9. Figure 9: The cross-dataset evaluation results. We evaluate the CoLa trained with historical documents on OBCs, Japanese characters and Korean characters. The left panels are character decomposition results, and the right panels are top-10 similar images of the examples. 21 [PI…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

49 extracted references · 42 canonical work pages

  1. [1]

    Shu, H., R. C. Anderson. Role of radical awareness in the character and word acquisition of chinese children. Reading Research Quarterly, 32(1):78–89, 1997

    work page 1997

  2. [2]

    Chan, L., T. Nunes. Children’s understanding of the formal and functional characteristics of written chinese. Applied psycholinguistics, 19(1):115–131, 1998

    work page 1998

  3. [3]

    Shu, H., R. C. Anderson, N. Wu. Phonetic awareness: Knowledge of orthography–phonology relationships in the character acquisition of chinese children. Journal of Educational Psychology, 92(1):56, 2000

    work page 2000

  4. [4]

    Shu, H., X. Chen, R. C. Anderson, et al. Properties of school chinese: Implications for learning to read. Child development, 74(1):27–47, 2003

    work page 2003

  5. [5]

    Tang, M., R. Liu, F. Xu. Compositionality in chinese characters: Evidence from english-speaking children. In Proceedings of the Annual Meeting of the Cognitive Science Society, vol. 46. 2024

    work page 2024

  6. [6]

    Schyns, P. G., R. L. Goldstone, J.-P. Thibaut. The development of features in object concepts. Behavioral and brain Sciences, 21(1):1–17, 1998

    work page 1998

  7. [7]

    Winston, P. H., B. Horn. The psychology of computer vision, vol. 67. McGraw-Hill New York, 1975

    work page 1975

  8. [8]

    Smith, L. B., S. S. Jones, B. Landau, et al. Object name learning provides on-the-job training for attention. Psychological science, 13(1):13–19, 2002

    work page 2002

Show all 49 references
  1. [9]

    Freyd, J. J. Representing the dynamics of a static form. Memory & cognition, 11(4):342–346, 1983

    1983

  2. [10]

    Lake, B. M., R. Salakhutdinov, J. B. Tenenbaum. Human-level concept learning through probabilistic program induction. Science, 350(6266):1332–1338, 2015

    2015

  3. [11]

    Salakhutdinov, J

    Lake, B., R. Salakhutdinov, J. Gross, et al. One shot learning of simple visual concepts. In Proceedings of the annual meeting of the cognitive science society, vol. 33. 2011

    2011

  4. [12]

    Lake, B. M., T. D. Ullman, J. B. Tenenbaum, et al. Building machines that learn and think like people. Behavioral and brain sciences, 40:e253, 2017

    2017

  5. [13]

    Wang, T., Z. Xie, Z. Li, et al. Radical aggregation network for few-shot offline handwritten chinese character recognition. Pattern Recognition Letters, 125:821–827, 2019

    2019

  6. [14]

    Zhang, J

    Wang, W., J. Zhang, J. Du, et al. Denseran for offline handwritten chinese character recognition. In 2018 16th International Conference on Frontiers in Handwriting Recognition (ICFHR), pages 104–109. IEEE, 2018

    2018

  7. [15]

    Chen, J., B. Li, X. Xue. Zero-shot chinese character recognition with stroke-level decomposition. arXiv preprint arXiv:2106.11613, 2021

    2021 arXiv

  8. [16]

    Radford, A., J. W. Kim, C. Hallacy, et al. Learning transferable visual models from natural language supervision. In International conference on machine learning , pages 8748–8763. PMLR, 2021

    2021

  9. [17]

    Yu, H., X. Wang, B. Li, et al. Chinese text recognition with a pre-trained clip-like model through image-ids aligning. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 11943–11952. 2023

    2023

  10. [18]

    Jin, L.-W., J.-X. Yin, X. Gao, et al. Study of several directional feature extraction methods with local elastic meshing technology for hccr. In Proceedings of the Sixth Int. Conference for Young Computer Scientist, pages 232–236. 2001. 10

    2001

  11. [19]

    Su, Y .-M., J.-F. Wang. A novel stroke extraction method for chinese characters using gabor filters. Pattern Recognition, 36(3):635–647, 2003

    2003

  12. [20]

    Techniques for solving the large-scale classification problem in chinese handwriting recognition

    Chang, F. Techniques for solving the large-scale classification problem in chinese handwriting recognition. In Summit on Arabic and Chinese Handwriting Recognition , pages 161–169. Springer, 2006

    2006

  13. [21]

    Cire¸ san, D., U. Meier. Multi-column deep neural networks for offline handwritten chinese character classification. In 2015 international joint conference on neural networks (IJCNN) , pages 1–6. IEEE, 2015

    2015

  14. [22]

    Wang, X.-Y

    Yin, F., Q.-F. Wang, X.-Y . Zhang, et al. Icdar 2013 chinese handwriting recognition competition. In 2013 12th international conference on document analysis and recognition, pages 1464–1470. IEEE, 2013

    2013

  15. [23]

    Huang, G., Z. Liu, L. Van Der Maaten, et al. Densely connected convolutional networks. In Proceedings of the IEEE conference on computer vision and pattern recognition , pages 4700–4708. 2017

    2017

  16. [24]

    Cao, Z., J. Lu, S. Cui, et al. Zero-shot handwritten chinese character recognition with hierarchi- cal decomposition embedding. Pattern Recognition, 107:107488, 2020

    2020

  17. [25]

    Kim, J. W., K. I. Kim, B. J. Choi, et al. Decomposition of chinese character into strokes using mathematical morphology. Pattern Recognition Letters, 20(3):285–292, 1999

    1999

  18. [26]

    Liu, C.-L., I.-J. Kim, J. H. Kim. Model-based stroke extraction and matching for handwritten chinese character recognition. Pattern Recognition, 34(12):2339–2352, 2001

    2001

  19. [27]

    Lin, Z., Y .-F. Wu, S. V . Peri, et al. Space: Unsupervised object-oriented scene representation via spatial attention and decomposition. arXiv preprint arXiv:2001.02407, 2020

    2001 arXiv

  20. [28]

    Jiang, J., S. Ahn. Generative neurosymbolic machines. Advances in Neural Information Processing Systems, 33:12572–12582, 2020

    2020

  21. [29]

    Van Steenkiste, J

    Greff, K., S. Van Steenkiste, J. Schmidhuber. Neural expectation maximization. Advances in Neural Information Processing Systems, 30, 2017

    2017

  22. [30]

    Greff, K., R. L. Kaufman, R. Kabra, et al. Multi-object representation learning with iterative variational inference. In International Conference on Machine Learning , pages 2424–2433. PMLR, 2019

    2019

  23. [31]

    Emami, P., P. He, S. Ranka, et al. Efficient iterative amortized inference for learning symmetric and disentangled multi-object representations. In International Conference on Machine Learning, pages 2970–2981. PMLR, 2021

    2021

  24. [32]

    Weissenborn, T

    Locatello, F., D. Weissenborn, T. Unterthiner, et al. Object-centric learning with slot attention. Advances in Neural Information Processing Systems, 33:11525–11538, 2020

    2020

  25. [33]

    Seitzer, M., M. Horn, A. Zadaianchuk, et al. Bridging the gap to real-world object-centric learning. arXiv preprint arXiv:2209.14860, 2022

    2022 arXiv

  26. [34]

    Mahendran, S

    Elsayed, G., A. Mahendran, S. van Steenkiste, et al. Savi++: Towards end-to-end object- centric learning from real-world videos. Advances in Neural Information Processing Systems, 35:28940–28954, 2022

    2022

  27. [35]

    Singh, G., Y .-F. Wu, S. Ahn. Simple unsupervised object-centric learning for complex and naturalistic videos. Advances in Neural Information Processing Systems, 35:18181–18196, 2022

    2022

  28. [36]

    Singh, G., F. Deng, S. Ahn. Illiterate dall-e learns to compose.arXiv preprint arXiv:2110.11405, 2021

    2021 arXiv

  29. [37]

    Kingma, D. P., M. Welling. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114, 2013. 11

    2013 arXiv

  30. [38]

    Sohn, K., H. Lee, X. Yan. Learning structured output representation using deep conditional generative models. Advances in neural information processing systems, 28, 2015

    2015

  31. [39]

    Darcet, T

    Oquab, M., T. Darcet, T. Moutakanni, et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193, 2023

    2023 arXiv

  32. [40]

    Matthey, C

    Watters, N., L. Matthey, C. P. Burgess, et al. Spatial broadcast decoder: A simple architecture for learning disentangled representations in vaes. arXiv preprint arXiv:1901.07017, 2019

    1901 arXiv

  33. [41]

    Yin, D.-H

    Liu, C.-L., F. Yin, D.-H. Wang, et al. Online and offline handwritten chinese character recogni- tion: benchmarking on new databases. Pattern Recognition, 46(1):155–162, 2013

    2013

  34. [42]

    Kingma, D. P., J. Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014

    2014 arXiv

  35. [43]

    Zu, X., H. Yu, B. Li, et al. Chinese character recognition with augmented character profile matching. In Proceedings of the 30th ACM International Conference on Multimedia , pages 6094–6102. 2022

    2022

  36. [44]

    Luo, G.-F., D.-H. Wang, X. Du, et al. Self-information of radicals: A new clue for zero-shot chinese character recognition. Pattern Recognition, 140:109598, 2023

    2023

  37. [45]

    Huang, D

    Li, Z., Y . Huang, D. Peng, et al. Sidenet: Learning representations from interactive side information for zero-shot chinese character recognition. Pattern Recognition, 148:110208, 2024

    2024

  38. [46]

    Zhang, Y ., Y . Zhu, D. Peng, et al. Hiercode: A lightweight hierarchical codebook for zero-shot chinese text recognition. Pattern Recognition, 158:110963, 2025

    2025

  39. [47]

    Yu, H., J. Chen, B. Li, et al. Chinese character recognition with radical-structured stroke trees. Machine Learning, 113(6):3807–3827, 2024

    2024

  40. [48]

    Li, Z., Q. Wu, Y . Xiao, et al. Deep matching network for handwritten chinese character recognition. Pattern Recognition, 107:107471, 2020

    2020

  41. [49]

    Zhang, X

    Wang, P., K. Zhang, X. Wang, et al. An open dataset for oracle bone character recognition and decipherment. Scientific Data, 11(1):976, 2024. 12 A Proofs and Derivations A.1 ELBO According to the stochastic gradient variational Bayes [? ? ], the log-likelihood p(F , y|X, ϵ, T ...

    2024

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.